Page 157 - Algorithm Collections for Digital Signal Processing Applications using MATLAB

146 Chapter 4

FIR Filter block as described above can also be replaced with the Back
propagation Neural Network block as mentioned in the chapter 1 to perform
filtering operation as described below. In this experiment the reference
signal and its corresponding corrupted signal are assumed to be known.

3.1 Approach

Step 1: Back propagation neural network (see chapter 1) is constructed with
11 input Neurons and 1 output neuron and 5 Hidden neurons (say)

Step 2: In signal processing point of view, input of the neural network is the
corrupted signal and the output of the neural network is the filtered
signal.

Step 3: During the training stage, the elements of the Input vectors are the
samples collected from the corrupted reference signal. Similarly
element of the output vector is the corresponding sample collected
from the reference signal.

Figure 4-6. BPNN Filter Structure